What it does Models.dev is a structured database of AI model metadata—pricing, context limits, modalities, knowledge cutoffs, and more—stored as human-readable TOML files in a GitHub repo. A GitHub Action validates submissions against a TypeScript schema, and the whole thing gets served as a JSON API at models.dev/api.json. Think of it as a community-edited spreadsheet that happens to be queryable.

The interesting bit The project is built to slot directly into the Vercel AI SDK ecosystem: model IDs match AI SDK conventions, and the npm field in each provider config points to the corresponding SDK package. There’s even an extends mechanism so wrapper providers can inherit from canonical definitions instead of duplicating data—a rare touch of schema inheritance in what could have been a flat key-value mess.

Key highlights

• Data lives as TOML files organized by providers/<provider>/models/<model>.toml, with SVG logos alongside
• JSON API and logo CDN (models.dev/logos/{provider}.svg) are auto-generated from the repo contents
• Schema covers costs per million tokens, context/input/output limits, modalities, reasoning support, tool calling, and release status (alpha/beta/deprecated)
• extends field lets non-first-party wrappers inherit and override canonical model definitions
• Validation via GitHub Action plus a bun run compare:migrations diff tool for safe refactoring
• Used internally by opencode.ai, so the maintainers are dogfooding it

Caveats

• The README notes they “need your help keeping the data up to date”—this is only as current as community PRs
• No explicit coverage of how many models or providers are currently tracked; you’d have to browse the providers/ directory to find out

Verdict Worth bookmarking if you build tools on top of the Vercel AI SDK or need to compare model pricing programmatically. Less useful if you just want a pretty UI to browse models—this is infrastructure, not a consumer app.